EXTENDED ABSTRACT: Flame-retardant polymer materials play a crucial role in national economic development and defense, with applications in building materials, aerospace, rail transportation, and military equipment. These materials are essential for ensuring public safety, industrial safety, and national defense security. However, current research on flameretardant polymers predominantly relies on trial-and-error methods, guided by flame-retardant mechanisms and experiences from non-real fire environments. This approach is inefficient and often results in unsuccessful design outcomes. Data-driven development, characterized by the integration of "big data + AI," has emerged as the fourth paradigm in material science, offering new tools for the high-throughput design of flame-retardant materials. However, these tools are effective only within specific systems, as models tend to perform poorly when applied to different material systems. This limitation stems from
the fact that existing flame-retardant research is often conducted in non-real fire environments or only collects physical and chemical properties during combustion. As a result, the data quality is low, the volume is insufficient, and there is a lack of a solid data foundation for developing high-accuracy flame-retardant models using artificial intelligence methods. To address this issue, this paper introduces a detection and analysis method that simultaneously acquires extensive data—such as heat release, smoke release, transient free radicals, functional groups, fine chemical structures, and collected smoke particles—within a single combustion experiment. We also develop a corresponding combustion analysis instrument. This data is then used to build a machine learning model to discover new flame-retardant materials. The proposed high-throughput burning experiment method provides a robust data foundation, facilitating a transition from trial-and-error research to a scientificdesign paradigm based on artificial intelligence models.
Keywords：flame retardant materials; high throughput detection; real-time online; combustion behavior